AI-assisted coding promises impressive results at remarkable speed, but getting a working first version is only the beginning. In this interview from the Software Architecture Gathering 2025, technical coach and software engineer Clare Sudbery explores what happens when LLM-generated code needs to evolve, be maintained, and meet real production standards. Drawing on more than 25 years of software engineering experience and her expertise in Test-Driven Development (TDD), she explains why established engineering practices can become even more important when developers hand part of the coding process over to an LLM.

Key Takeaways:

1. TDD provides an essential way to verify the quality and behavior of LLM-generated code.
2. AI-assisted development makes strong software engineering practices more important, not less.
3. LLMs can accelerate the first steps of a task without necessarily making the entire development process faster.
4. AI can amplify both good and bad coding practices.
5. The benefits of AI need to be weighed against concerns around security, copyright, environmental impact, and human consequences.

Clare Sudbery offers a nuanced perspective on AI-assisted software development that goes beyond both enthusiasm and rejection. While LLMs can be useful for overcoming an initial hurdle, exploring unfamiliar technologies, or bootstrapping configurations, she argues that their value depends heavily on the practices surrounding them. TDD provides one way to retain confidence and control as more code is generated by machines. At the same time, questions around security, copyright, environmental impact, and the broader human consequences of AI remain impossible to ignore. Watch the full interview below to hear why Clare is still “on the fence” and why that might be a valuable position to take.